Selecting for intelligence feels likely to be banned (unfortunately as this could be hugely beneficial for humanity if widespread) but in theory selecting against disease should also be pro intelligence as both are correlated through mutational load, right?
Given the prices involved here and methods used, this might be...really difficult to ban? If you let parents see each embryo's data before they make a choice, and the information necessary to make that selection is on SciHub, then, uh, *wink wink*...
Would be hard to ban any use of this method considering how many diseases it can fight (and how sympathetic people suffering from these diseases are), and I think any use of this method quickly results in selection for whatever parents want (whether that's intelligence or aesthetic traits).
Yeah, I agree. In fact, I don't know if you even need the data. Here's an article I saw linked over at marginal revolution a while back:
"Children resulting from frozen embryos were more socially adept than those implanted fresh after eggs were fertilised. The children also moved better, had superior communication skills and showed more independence. Allan Pacey, a fertility expert at the University of Sheffield, suggested that this was caused in part by the rigours of the thawing process. Not all embryos survive thawing, and perhaps those that do are “stronger”, he said.[11] Perhaps freezing and thawing embryos is an inadvertent eugenics process, most successful with embryos having a low mutation load."
There does seem to be a selection process, post implantation.
"Good obstetric outcome (GBO), defined as a singleton, term, live birth with appropriate for gestational age birth weight, was the primary outcome measure. Secondary outcomes included live birth, clinical pregnancy, spontaneous abortion, preterm birth, multiple births and gestational age-adjusted weight. Outcomes were modeled using the generalized estimating equation approach.
MAIN RESULTS AND THE ROLE OF CHANCE
"Overall, fresh transfer was more likely to result in a live birth (55.7% versus 39.5%; adjusted risk ratio 1.21; 95% CI 1.18–1.26; P < 0.001). "
Well, I'm not sure they're not entirely "opposite" results. If freezing meant embryos with problems were less likely to develop to term that would explain the improved health outcomes from frozen infants.
I'm saying that in contrast or addition to the Marginal Revolution advanced hypothesis "Not all embryos survive thawing" it could be that thawed embryos with health issues are less likely to be delivered to term.
If the observations are all accurate then I really wonder what mechanism mediates this.
That makes me wonder-- is freezing merely a mechanism to improve outcomes for IVF, or can it be used to improve outcomes even relative to non-IVF natural conception?
That is, would it make sense for couples to freeze embryos specifically to filter out worse ones (aside from the usage of PGS)?
If it's banned in the US but legal in, say, Brazil, then presumably a lot of wealthy ambitious couples will go on a long vacation (or several short ones) in Brazil and come back with the woman pregnant, and regulators in the US unable to do anything about it.
Citizen jurisdiction. They might not be able to force the woman to have an abortion ("cruel and unusual"), but a country can absolutely throw its citizens in prison for what they did abroad.
(The USA in particular also has some really dodgy long-arm statutes which give them jurisdiction if any money touched the US banking system at any point.)
I hope the Overton window shifts enough that selecting for intelligence gets accepted -- I always think of this passage by Nick Bostrom in https://www.nickbostrom.com/views/science.pdf:
"There are three ways to contribute to scientific progress. The direct way is to conduct a good scientific study and publish the results. The indirect way is to help others make a direct contribution. ... A third approach is to marry the first two and make a scientific advance that itself expedites scientific advances.
A “superficial” contribution that facilitates work across a wide range of domains can be worth much more than a relatively “profound” contribution limited to one narrow field, just as a lake can contain a lot more water than a well, even if the well is deeper.
No contribution would be more generally applicable than one that improves the performance of the human brain. Much more effort ought to be devoted to the development of techniques for cognitive enhancement, be they drugs to improve concentration, mental energy, and memory, or nutritional enrichments of infant formula to optimize brain development.
Imagine a researcher invented an inexpensive drug which was completely safe and which improved all‐round cognitive performance by just 1%. The gain would hardly be noticeable in a single individual. But if the 10 million scientists in the world all benefited from the drug the inventor would increase the rate of scientific progress by roughly the same amount as adding 100,000 new scientists. Each year the invention would amount to an indirect contribution equal to 100,000 times what the average scientist contributes. Even an Einstein or a Darwin at the peak of their powers could not make such a great impact. Meanwhile others too could benefit from being able to think better, including engineers, school children, accountants, and politicians."
But interventions like salt iodization only help better realize potential, not increase it. It's the latter I'm curious about. Unfortunately this is a really really hard problem: https://www.gwern.net/Drug-heuristics
Practically, what's the difference between "realizing" potential and "increasing" it? Is it the upper bound of what's possible? I don't know if we know what that upper bound looks like for polygenic testing for either a given individual or a population.
Yes, the main difference is about the upper bound. Addressing iodine deficiencies (from the limited reading I've done) is likely to be a very cost-effective method to improve cognitive ability in the developing world. I think it's a shame we don't prioritize these types of low-cost interventions that have such huge long-term effects.
In regards to your second question, the upper bound for polygenic testing for a single generation is about 1 standard deviation for any given trait given current limitations on oocyte production and GWAS sample sizes. For most diseases it's closer to 0.3-0.5 standard deviations.
For multiple generations the limits are much much higher. For example, Steven Hsu has estimated that if you naively add up all the alleles that contribute positively to IQ, additive variance alone would predict that genome would have an IQ of about 1000. Obviously in practice that wouldn't work even if you could do it because at some point the relationship between positive alleles and phenotype would break down. But it gives you an idea of what might be possible in the future.
Further evidence can be found in animal and crop breeding programs that have been going on for thousands of years. We have increased the oil and protein content of the ancestors of modern corn by 30 standard deviations over the last 10,000 years. And while humans are not corn, many of the genetically influenced traits we care about (like disease risk) are controlled by a similarly large number of genes (meaning they are probably capable of undergoing similarly large increases with enough selection pressure).
"For example, Steven Hsu has estimated that if you naively add up all the alleles that contribute positively to IQ, additive variance alone would predict that genome would have an IQ of about 1000."
We seem to agree that that wouldn't actually work. So doesn't that indicate a problem with the model we're using? We do agree that not all improvements are additive, right? Some may just be different ways of accomplishing the same thing.
I feel like it's hard to predict whether this technique could produce people who were that much faster, better, or stronger than the best currently existing humans without some kind of tradeoff. Eventually we'd presumably start running into issues related to cranial capacity. Or trading off generalized for specialized intelligence.
The notion of "+.2 IQ points for this gene with no consequences or costs" just seems deceptively ... neat and predictable.
"Further evidence can be found in animal and crop breeding programs that have been going on for thousands of years. "
Sure. And the Green Revolution has accelerated that improvement with fertilizers, pesticides, watering, etc. It's also produced plants that don't survive well outside of a farm. It's produced plants that serve humans rather than themselves (since obviously wild plants have had millions of years to improve themselves for their own purposes.) Whats the equivalent of applying that metaphor to humans? Being able to use more fertilizer and water, as provided by farmers, is win-win. Maybe human intelligence can do something similar. More food = more cholesterol for neurogenesis? But then, we've demonized cholesterol and we give people drugs so they produce less of it so they can live longer. At some point crop improvements are going to have to hit some kind of physical limit in terms of photosynthetic efficiency and available light. Even if per hectare crop production is not yet leveling off for some crops. Monoculture offers a benefit in terms of being able to harvest ripe crops at a set time, reducing poaching by animals. But the loss of genetic diversity also increases vulnerability to pathogens. Genetic diversity exists for a reason. IQ positively correlates with alcoholism, for example.
An endurance runner has more slow twitch muscles. A sprinter has more fast twitch muscles. Once we eliminate various physiological defects, further improvements are tradeoffs. And I don't know how to be certain that this wouldn't be an issue with polygenic testing.
Polygenic testing is a very exciting development. It can help address things like increasing mutational load in an ethical fashion. But we're dealing with self modifying systems whose behavior we can't fully predict or understand. That seems like a setup for surprises.
I think talking about Overton windows is rather premature at this point, as there's no hard evidence that embryos can be effectively screened for intelligence at all. Merely assigning a numeric score based on a bunch of markers isn't very reliable for an extremely polygenic trait such as intelligence... especially when there's still active debate going on regarding what proportion of IQ is genetic; and even to what extent IQ correlates to real-world outcomes; or whether it even makes sense to use a single scalar quantity to answer such questions.
It’s 85 percent heritable (g), not iq scores. Especially at adult-age. Less suppressed variation due to being forced to go to school or parental choices. Cumulative exponential effects from learning faster and gaining skills faster. Same reason why bad diet in NKorea attenuates and reduces intragroup variation. Phenotypic variation is maximized and potential is realized when opportunities are maximized; sex-linked behaviours, career choices, intelligence, anything. Breeding dogs and animals have exact morphological similarities as hybridization of haplotypes or human breeding groups, it’s just politicized and suppressed because the repercussions are too big to be accepted (non-conscientious, psychopathic, low intelligence being a feedforward loop of more undesirable societal outcomes and offspring). And au contraries to the many individual variant effect hypothesis, it’s mainly few main progenitor genes with integrated mechanisms that are deterministic of neurological/biophysical health that are conserved with additional improvements made by a combination of rare and minute pathways that increase the efficiency/differentiation of brain tissue/change of rate of growth and thickening which increases the magnitude of intelligence gains. Not just a series of 0.00012 percentile variants that improve myelination or conductivity. dendrite dispersibility. Scalar measurements are invariant with respect to any and all cognitive tasks. More instructions, more complexity is inversely proportional to success rates; that includes taking medicine, making mistakes, accidents or anything. Statistical artefacts are also irrelevant and impertinent. China will be dominating and probably Israel too. The elite population has already been practicing eugenics discretely with IVF reiterated embryo selection and cross marriages stratified only by high conscientiousness, agreeableness, intellect and desire for dominance.
Judaism is bioethically permissive (abortion is legal in Israel, no one cares) and there isn't much wokeness, and there is a national defense reason to add +10 IQ to everyone so I expect them to be first or among the first to do embryo selection.
You don't have to understand all of the genes in order to select for higher intelligence. It may be extremely polygenetic, but we can still identify the gene variants with the largest effects and select for those. I think the Overton window is very relevant, since we're going to have the technology to attempt it long before we know whether it's actually possible.
To know if we can select for intelligence, we'd basically need to just do it on a massive scale in humans, then compare the IQs of children selected for intelligence to children not selected for intelligence (and obviously account for the many, many confounders). The question is whether we allow this experiment to take place or not.
> It may be extremely polygenetic, but we can still identify the gene variants with the largest effects and select for those.
As of today, we really can't do that, because intelligence is not just a linear combination of a bunch of genes. Like any polygenic trait, it's a complex network of interactions between genes and regulatory regions, all of which upregulate something while downregulating something at the same time. So assigning an IQ score based on a bunch of markers is not a simple matter of multiplying their association scores together. It's a search in multi-dimensional space, and the problem with brute statistical models -- yes, even ML models -- is that if you have enough dimensions, you have enough degrees of freedom to pretty much find anything you want.
As I said in my comment above, if someone had a model that provably selects embryos who then grow up into children who reliably perform better in the real world than controls, then I would be impressed. But merely throwing GWAS at the wall to see if it sticks is less interesting to me.
> Like any polygenic trait, it's a complex network of interactions between genes and regulatory regions, all of which upregulate something while downregulating something at the same time.
It's true of course that the genetics of intelligence are complicated. But you're incorrect in saying that this makes it impossible to select for. Most of the variance in highly polygenic traits can be explained with linear effect models. See this paper for a more thorough analysis: https://www.biorxiv.org/content/10.1101/2020.02.12.946608v1.full
> if you have enough dimensions, you have enough degrees of freedom to pretty much find anything you want.
Only if you use a weak statistical threshold for significance. But most modern genome-wide association studies use a very strict significance threshold of 5*10^-8, at which point you don't get many false positives.
> But merely throwing GWAS at the wall to see if it sticks is less interesting to me.
This is not an accurate description of how polygenic screening is done. All of the predictors used by companies currently doing pre-implantation screening are validated using test sets of siblings that weren't in the training data. And while the predictors used today don't capture all genetically caused variance (usually they capture around 20% of it), they do significantly better than chance at picking out the sibling who will go on to avoid developing a tested disease.
There are some limitations to these results (for example predictors of disease risk created with a training set of people with mostly European ancestry don't generalize well to non-Europeans), these limitations are essentially due to differences in the frequencies of certain alleles among populations whose common ancestors are sufficiently distant.
This in turn is caused by something called "linkage disequilibrium". When DNA from a sperm and egg crosses over during meiosis, the locations of such cross-overs are essentially random. But more often than not, base pairs on the chromosomes that are physically very close to one another on the genome will remain together after crossing over. This creates a problem for geneticists: if you're trying to figure out which letter in a protein-coding region is causing a 2% increase in risk of breast cancer, how do you know whether it's the A or the T right next to it? They're almost always inherited together due to linkage disequilibrium.
For people with recent ancestors, this doesn't have much of an effect on the prediction accuracy of these GWAS tests; the A and the T are very likely to be inherited together, so no matter which one was causing the increased risk of Breast Cancer, selecting against one will probably select against the other. But the longer ago two people last shared a common ancestor, the more likely it is that one of their ancestors had a split between those the letter that actually causes the change and the one that you've pinpointed as most likely to cause it. If that happens, your predictor becomes useless because it's incorrectly assigning causation.
Fortunately there's a solution. In the short term with the relatively small sample sizes we have today, we can restrict predictor use to populations in which they have been validated to work well. And in the longer run, GWAS need to start including genetic/phenotype data from a much larger, more genetically diverse group of people. Doing so will both improve the predictors AND make the benefits of embryo screening available to a larger group of people. And we're likely to develop many useful screening tests along the way that can be used to target health interventions at people most at-risk for certain conditions.
I don't think something has to be achievable for there to be a productive discussion about Overton windows. Five years ago self-driving cars were a novelty that still to this day don't function well enough for commercial use, but there were all sorts of ethical discussions around their decision making that helped create a societal opinion about the technology. Similar discussions take place all the time with the potential problems/benefits of AI and how we should approach the technology.
Boosting human intelligence through this type of gene selection is a bit further off than some of these things, but the potential is there for it to arrive within our lifetimes, and discussions about it will surely happen before it is fully feasible.
>Five years ago self-driving cars were a novelty that still to this day don't function well enough for commercial use, but there were all sorts of ethical discussions around their decision making that helped create a societal opinion about the technology.
I'm pretty sure that the "societal opinion about the technology" is that society expects the technology to produce cars that simply do not crash, and that nobody who matters is going to say "It's great that our cars make what we consider proper ethical decisions about who or what to crash into". The consequentialist nerds got to imagine that the Trolley Problem finally had a real-world application; everybody else said "if you're talking about what the car will decide to crash into, get back to us when you've got the car deciding not to crash".
We already know polygenic scores can predict which sibling has higher intelligence and educational achievement. Predicting which embryo will is no different.
If having 100,000 more scientists made a huge difference, we would *already* have seen that effect, since there are *way* more scientists working today than at any time in human history, even on a per capita basis. If it made a giant difference, we ought to be seeing Einstein/Fermi/Newton breakthroughs on an annual basis, certainly much faster than we saw them before 1939 -- and nothing could be further from the truth.
The entire calculation is based on a dubious premise about the nature of scientific advance, id est that it proceeds like some kind of construction of an earthen dam, and if you have 200,000 peasants with buckets it gets done 2x as fast as if you had only 100,000. But so far as I can see, an examination of history suggests quite the contrary, that science advances largely by abrupt and difficult to explain (let alone predict) highly contingent paradigm shifts that are the result of some lucky coincidence of improved instrumentation, cultural readiness for a paradigm shift, and some well-prepared brilliant mind that happens to be born at the right time in the right place. You might say it happens by quantum transitions, like absorption of an visible photon to cross a band-gap, and in such a case it doesn't really matter how many IR photons with energy much less than the band gap you provide, nothing happens until the big hv comes along.
> If having 100,000 more scientists made a huge difference
I think you're thinking of "huge" in relative terms, while Bostrom (and I) think of it in absolute terms. Like how a billion dollars is a huge amount of money, but negligible vs say Apple's market cap. I agree in relative terms it's small, especially given that (to quote Scott) "constant growth rates in response to exponentially increasing inputs is the null hypothesis" (https://www.lesswrong.com/posts/v7c47vjta3mavY3QC/is-science-slowing-down). It's still a lot larger than the first 2 sorts of advances, usually.
I agree with the content of your second paragraph, but am puzzled by how it wouldn't be affected by talent pool size (whether size = number of people or "amount of talent", eliding difficulties in aggregating talent for this to be measurable).
It's affected by the talent pool size up to the limit of saturation, but not afterward. Think of enzyme kinetics: once you saturated the enzyme, you don't get any more product by boosting the substrate concentration, because the substrate concentration is no longer the limiting factor on the rate.
So the question is: have we reached saturation on the talent pool yet? That is, is everyone who *can* do scientific breakthrough thinking identified, recruited, and remunerated sufficiently that the availability of talent is no longer the limiting factor? I think a decent case can be made for "mostly yes." You don't really see much in the way of Ramanujans escaping otherwise impoverished lives by sheer luck (writing a letter to Hardy). We're pretty darn good at identifying people who have the requisite smarts and easing their way into the field if they want to be in it. (With some caveats, of course: I'm amused that the woman who was one of the pair that invented the modified-nucleoside tech used to make the mRNA COVID vaccines had a rough time at Penn and ended up leaving because people thought she was an unproductive crank, and now there are people calling for her to share the Nobel for the work.)
That begs the question of what the limiting factor might be, if it isn't the size of the talent pool. As I said below, I don't have any strong ideas on what it is, other than some vague feeling that resources are not quite distributed right, that the *most* talented of the talent pool are getting the most time/money to do their work, a the expense of the less talented. (And here I mean "talented at producing breakthrough ideas.") But I have no idea how to identify such people algorithmically, although I tend to feel I can spot them when I meet them -- it's a certain weird independence of thought, an ability to spot the weakness in the conventional wisdom. It's the person who says "this phlogiston theory is all well and good, very logical and explains a lot, but I wonder what happens if I oxidize some mercury."
"A common theme across all of these is that fairly obvious opportunities were not pursued by incumbent institutions. We give examples of actions Fast Grants took not to indicate some kind of supposed brilliance but rather to emphasize the opposite: Fast Grants pursued low-hanging fruit and picked the most obvious bets. What was unusual about it was not any cleverness in coming up with smart things to fund, but just finding a mechanism for actually doing so. To us, this suggests that there are probably too few smart administrators in mainstream institutions trusted with flexible budgets that can be rapidly allocated without triggering significant red tape or committee-driven consensus."
For the sort of science that needs a research laboratory to be done properly, the limiting factor is probably the supply of research laboratories. Fully stocked, staffed, and maintained research laboratories are not cheap, and lots of people who aspire to be and have the education/credentials to be scientists seem to never land a position in one (except maybe as chief bottle-washer to a "real" scientist).
There's some science that can be done just by staring into a coffee cup and thinking real hard, https://www.beanthinking.org/?p=1019, or maybe poking at the keyboard of a commodity PC. But there's a great deal more science that deceptively seems like it should be this sort of science, but veers rapidly into junk if not grounded by experimental verification and model validation. For which, you need research labs, and if you exceed the research-lab limit you just wind up with lots of wannabe scientists thinking up junk science.
> I think a decent case can be made for "mostly yes."
I think it's "almost certainly not". The internal politics, incentives and funding structures around science are pretty perverse, and I think this drives a lot of talent away.
I would like to challenge your central premise: that we aren't seeing more scientific development than ever before.
Medicine is the classic example. Today there are people studying and creating scientific breakthroughs on rare diseases. We lay people don't see this unless we have a relation struck by this rare condition. These are really hard, breakthrough discoveries that save lives. Are they society changing? Yes, if it's your loved one.
Yes we haven't learned how to increase our rate of Einsteins (not sure I agree but I'll accept it), but the breadth across disciplines and niches that we are advancing human knowledge today is unprecedented.
I'm not sure, but I don't think I agree with that. If we consider advances in the context of what already exists, and the resources available for implementation, I think a strong case can be made that there are many periods of history in which general technological advance was significantly higher -- and was reflected in a much strong rate of general social improvement. Let's say the advances in chemistry, thermodynamics, engineering, metallurgy and civil engineering in the 1600-1700 timeframe, for example. Or even the advances in physics in the 1880-1950 timeframe.
I agree biochemistry or molecular biology is having a renaissance right now, but I don't agree that's due to the number of people working in it. Those advances are largely built on factors that were invented 50 years ago -- breakthroughs in instrumentation (NMR) or technique (PCR), or brilliant insights (Watson-Crick model). The surge of people into the field in the last 20 years is I think a "Gold Rush" like *consequence* of smart people realizing the intellectual endeavors in this field might pay off better than in others, rather than the reverse.
I've been in the business (science research) a long time, and my anecdotal observation (insert the usual caveats) is that advance really is a lot like spectroscopy in quantum world: to go from Here to There, where There is some nontrivial and significant advance, it usually seems a quantum event, someone has to just take one big jump, all at once, and it simply can't be duplicated by any number of people taking any number of small jumps.
There are far more scientific papers published today than were published when I started in the business in the 1980s, but I would be hard-pressed to say that the density of genuinely transformative ideas has increased. We may have essentially hit "saturation" in the sense that throwing more (random) people and money into the process won't get us anything but more chaff, not any more wheat. It's not really the rate-limiting step any more.
What is, though? I don't know. I would vaguely say "getting money/time into the *right* hands -- of people who are capable of genuine breakthrough thinking." Problem is, they're really hard to identify. The guy who invented PCR is basically a surfer bum. There are plenty of people who had one brilliant idea and did zip before and after. It seems deeply serendipitous in many (althought not all) cases. How you improve this I don't know, although you might be able to nibble around the edges by improving the ability of people to just sit around and tinker with "dumb" (by the judgment of committees) ideas, since it's really only ideas considered at the time by the majority to be "dumb" that can turn out (in hindsight) to have been brilliant.
I can think of an avenue that isn't mentioned here but that I'd consider pretty damn important: academia is a horrid place to work at. Researchers have little choice in what they study (and I'd argue that most of the greatest advancements came from people obsessed with their field of study, whose research was guided by interest and their own ideas, not primarily where they'd find an underpaid job). Wages are laughably low. Job security is basically nonexistent unless you get tenure. If driven, intelligent, ambitious people don't get a chance to pursue their ideas, then they likely won't make the breakthrough they are capable of, and very few people really get that chance.
Anecdote: a friend of my mother's defended her thesis while undergoing chemotherapy, and lived apart from her husband and daughter for a decade because their careers were impossible to consolidate in a single city. She's a professor now, but nobody should have to go through that to have a shot at this.
How many people give up on research because of life circumstances? Not wanting to sacrifice their family, or sanity, for meager pay?
Then there's the ruthless pressure to publish, tweak data, the time spent writing grant applications etc, and the incentives that lead to large numbers of basically worthless papers, studies done on twenty college students and the like. Many of the people writing those want to do real research, but can't.
I don't have a solution here, but if we could aim the existing brainpower better (by which I mean, aim it less, because self-aimed talent is always more effective than people forced to work on stuff they're not passionate about) that might lead to more advances than increasing that brain power, or doubling the number of people possessing it.
All this (to the extent it holds) applies to fundamental science. On the margin, would additional researchers have more benefit in fundamental science (which is mostly done in academia) or at applied science (which is mostly done at private companies)?
Somebody wrote that quite some breakthroughs of past centuries were done by English clergy who had comfortable and safe living conditions, good education and enough time on their hands to pursuit whatever they liked. They had no hightech labs of course.
Pre-singularity UBI is probably going to be too diffuse for this; you need concentrations of wealth among smart well-educated wannabe researchers. And it works better if those people can hire lab techs, for which you need people who need jobs (or at least strongly benefit from jobs, which is hard to square with a UBI generous enough to buy lab equipment).
I assume that UBI gives people slack for crowd-funding and institutions, so at least small research projects involving multiple people should happen.
Those English clergy-- did them having real but not onerous day jobs do them any good?
UBI also relates to the argument from Bret Weinstein and Heather Heyer that universities are biased towards expensive research that they can take a cut of, and there isn't enough theoretical work getting funded.
That was a great read, thank you very much, though it doesn't come away saying "everything's fine", rather "It's shit, but it still works, somehow." Life (science), uh, finds a way.
I suspect that there's some overlap with the issue of hiring competent managers. Most bosses suck at bossing - you'd think this is something the market could figure out, but it doesn't. Maybe you could even link it to politicians. The underlying common problem is one of selection of leaders/elites and of effective allocation of limited resources, and market failures seem to be the norm, as well as failures of pretty much every other mechanism we know of. I expect smarter people than me have tried to tackle this, so I have nothing to contribute at this point.
> Selecting for intelligence feels likely to be banned
Quite possibly it will be in some jurisdictions. But if not all jurisdictions ban it, then there are going to be early adopters who want to do it, and once a significant number do, a lot of other people are going to want to, to keep up. There's also the point that if you have the raw data, they can't ban it.
Having said that, +3 IQ points is not a great deal, and for most prospective parents they could get a bigger boost than that with sperm/egg donation. Some people wouldn't want that, but others would.
You say that other people are going to want to select for intelligence to keep up with the others, but how many IVF babies are there going to be to keep up with, realistically? 1-2% of annual US births are IVF, and unless something happens to drastically increase that number, I don't see how adding 3 IQ points to 1-2% of the population (at most) is a compelling reason to unban a practice people consider unethical.
Adding 3 IQ points to the entire population may be easily worth the cost of conceiving every baby through IVF. IVF can have some complications, but they don't seem particularly serious or unsolvable. (You say "unban", but is it currently banned?)
I don't necessarily disagree that conceiving every baby through IVF would be a net positive, but how do you see that happening? I can't imagine how much it would cost to scale up IVF facilities and staff 100-fold, let alone the costs of actually operating them at that scale. Is that really worth the cost? It seems exceedingly unlikely that there aren't better ways to spend that money for the same, or more, benefit.
Things usually cost less, not more, when scaled up. (They might cost more if we are hitting resource limits, but that's unlikely to be a factor for IVF.) Assuming a generous $20,000 for the cost of IVF and genetic selection, it easily pays for itself through the lifetime of the child, even with modest assumptions about the correlation between intelligence and income. And that's not even accounting for the knock-on effects (such as smart people providing greater economic surplus to others in addition to making more money themselves, or a smarter society making better political decisions).
In a world where you can select for intelligence, you can likely select for a number of other health/well being factors as well. Anyone with the financial means to do so would then face pressure to use IVF, and a scale up could realistically happen I think.
That world is a ways away, if it ever arrives, but I don't think its outside the realm of real possibility.
Can we acknowledge somewhere that for all this IVF you need an awful lot of women who could conceive naturally to consent to it? I wouldn’t have put myself through that to raise my kid’s IQ by three points. Maybe if I had a sufficiently horrible disease running in my family, but the assumed chance of various common diseases we do tend to get wouldn’t have prompted me to do IVF just so I could do polygenic screening.
+3 IQ points is a bigger boost in cognitive ability for someone starting from a higher base. I suspect the boost is not uniform -- there are only so many SNPs you can optimize, and smarter folks are already closer to the optimal setting.
But thinking about IQ makes this boost seem zero-sum -- after all, the number itself is relative to the population, not an absolute measure of cognitive ability, i.e., not everyone can "win" in this IQ boosting game. But suppose there was some score for the aggregate effect of those SNPs. Then we'll have an absolute measure for how much headroom we all have (how far do we have before we've optimized all the SNPs?); perhaps all of us, including the smarter ones, are laughably far away from this human ideal. But against this absolute measure (and not a relative measure like IQ is) everyone can (in principle) win, and the ultimate victory condition for the human race is that everyone has maxed out their SNP score, and everyone starts from a level playing field.
>Selecting for intelligence feels likely to be banned
Does not feel likely to me.
The 'don't discriminate against/devalue people with low IQ' argument is something that gets heard a lot by people who fetishize IQ (us), because they represent the natural opponent 'side' to the type of person who would make the argument, and the social media ecosystem is efficient at throwing people into their natural opponents to produce content.
But I don't think any of this is something the average person thinks about, and I think the average person is pretty positive on the idea of IQ=good. I don't think it would occur to enough normal people that this might be something you would want to ban, certainly not enough to result in regulation.
Well in the US maybe. But some other country will allow it and own the world in a generation or two, and everyone in the US will [still] be angry and miserable.
If they're doing the whole thing for $1400, then they're not doing full sequencing on all the embryos (unless there have been some huge breakthroughs which I missed). They're probably using rna microprobes like 23&me does. This means the ability to select for traits they didn't have in mind will be limited.
No, they're certainly not sequencing the entire genome. They're just doing the same old looking for SNP fingerprints. That is presumably why the results are so poor.
Full sequencing is now much cheaper than $1400. Nebula Genomics is doing it right now for $499, Dante Labs runs yearly sales offering it for 300 euros etc., the price has really come down a lot in the last 2-3 years.
It's down to $299 for accurate full-genome sequencing at Nebula. So assuming it's $1400 + $300 per embryo, one cycle of 10 embryos will cost $12k for the basic IVF + $300*10 + $1400 = $16400
It's not linear so it sort of depends on where you are in the distribution. I'm assuming the average IVF-user will be around 115.
Assuming that $1500/year keeps up with inflation, and our discount rate equals the 30 year treasury rate equals the long term average rate of inflation, we can just multiply $1500 by about 40 years of working to get $60,000 as the value of one IQ point.
If instead your discount rate is 4% above inflation (a reasonable forward estimate of the expected return of SPY, which is below historical averages but justified by the abnormally high PE ratio presently) and the income starts 25 years after the IVF, then the value of an IQ point comes down to about $11k, and it's still worth it to get 3 points for $15200.
But assuming the $1500/year only keeps up with inflation is too pessimistic -- it should track overall income growth, which should be 2% faster than inflation. If SPY is 4% above inflation but only 2% above income growth, we use a 2% discount rate on the $1500/year annuity from 2045-2084 to get a present value of $25k.
25k is a lower bound on the value of one IQ point because it only considers the effect on individual income and not any of the other benefits of having higher IQ, such as better health and positive externalities on the community (through better science, better businesses, better political institutions, lower crime).
Getting 3 IQ points with a present value of at least 3x25k for only $16400 is an excellent deal.
The government taxes an average of 29% of people's incomes (https://www.thebalance.com/what-the-average-american-pays-in-taxes-4768594), so from the government's point of view it should be very worth it to pay $16400 now for 29% of $75,000 worth of future earnings. And arguably the government should use a lower discount rate than individuals because its opportunity cost is paying off 30 year treasuries yielding 2% instead of buying SPY yielding 6%. 29% of 3x60k is $52k of present value of future tax revenue that the government could get from paying $16400 for the whole package of IVF and embryo selection. (Before even counting the government's reduced healthcare costs due to the disease screening part).
I think that's probably not the full cost for this purpose.
(disclaimer - I've done some genome sequencing, but only of bacteria.)
The $299 nebula deal is for 30-fold coverage - that's an average, so some regions may be missing. I can't see which sequencing technology they're using, which affects what how much coverage you'd want, and also what sort of artifacts you'd expect to see. All these human sequencing packages only work because there's a good reference, so you might end up with reference sequence replacing a difference they couldn't get a good read of.
That only matters if that region actually differs, and you'll still catch most mutations I imagine. If you care about which alleles are together on the same chromosome, you'd really want to be using a long read technology. Using different chemistries together is good to get better sequences, but I doubt they're doing that.
Don't get me wrong, it's certainly impressive. Given that it's discounted from $1000, it's possible this price is a loss-leader using investor money to build the business, or something. But for the sake of argument, let's suppose not.
However, there's also the issue that an embryo isn't very many cells. You can afford to pinch a few for testing, but for a near-complete sequence you want at least many.
I'm not clear on how much saliva Nebula needs for that service, but it might be ... quite a lot of DNA. I found these statements on the web:
* "The median yield of DNA from a 2 ml saliva sample using Oragene is 110 µg", and
* "How much DNA does a human cell contain? A human cell contains about 6 pg of DNA.", and
* "DNA Sample Submission- Typically 100 to 1000 nanograms of DNA are required for whole genome or whole exome sequencing."
So realistically you probably want to aim for at least something like 200 pg of DNA as an input, or about 33 cells-worth - this ignores all losses in purification.
(You might think that 30 cells would be the minimum /anyway/, if you're getting 30-fold coverage. That's not necessarily the case, but if not it's probably also not complete coverage; you'd be sequencing some molecules multiple times and others none.) Anyway, I imagine these companies are using significantly more starting material than that.
So either you need to grow the cells up, or clone the DNA. Starting with one or two cells, the latter approach would probably give artifacts (missing sections, over-copied sections, sequence changes). If you care very much about the differences, you're going to have to go with the first option - which is going to cost more.
Aren't you assuming the IQ points are a certainty for the cost invested? That is, the investment is risk-free? But biology doesn't work that way. You're investing $15,000 for a certain probability distribution of outcomes with a mean of +3 IQ points. But there is certainly a range of outcomes, and what is the width of that distribution? If it's wide enough it's a pretty risky investment, since the odds are not bad that you get zip for your $$, and don't you need to do some discounting because of the increased risk?
I woudn't think so unless the value of the tails of the distribution is asymmetrical. For early adopters of this technology, I would actually expect it to work in the opposite direction: since the marginal benefit of one additional IQ point is steepest at the tails, and since the average early adopter of IVF is likely to be somewhat above average, the high tail of the distribution is likely to have more positive value than the lower tail, which is closer to average.
To be clear, does a couple doing this with a donor take up additional eggs that could have gone to somebody else, or does it use as many as "traditional" ivf? Neither this post nor the first linked article specified
Generally speaking intended parents will purchase an entire cycle’s worth of eggs from a donor and get the number they get. There are companies that pre-harvest the eggs and sell them one by one but it’s less common.
>Are these tribes based on geography? Are they based on race, ethnic origin, religion, IQ, what TV channels you watched as a kid? I don’t know.
>Some of it is certainly genetic – estimates of the genetic contribution to political association range from 0.4 to 0.6. Heritability of one’s attitudes toward gay rights range from 0.3 to 0.5, which hilariously is a little more heritable than homosexuality itself.
>(for an interesting attempt to break these down into more rigorous concepts like “traditionalism”, “authoritarianism”, and “in-group favoritism” and find the genetic loading for each see here. For an attempt to trace the specific genes involved, which mostly turn out to be NMDA receptors, see here)
I can think of a few governments that might want to select babies for personality, so that citizens are compliant and non-rebellious. And i can think of a few political-moral points of view that might also want to select babies on personality, so as to get future people more likely to agree with their point of view.
The big limiter here is "if you're doing IVF anyway." I doubt anyone would opt for IVF just for the benefit of genetic selection, and I don't think medical providers do the procedure unless natural conception is unsuccessful or inadvisable.
That could happen, but I think it would be many generations away. Unless almost the entire population was screened this way it would take an awful long time for genetic selection to weed this out, assuming people continue to intermarry between classes of non-IVF/IVF born humans.
Maybe far in the future, after all genetic risk factors are removed from the gene pool. But it seems unlikely to me that we'd abandon this practice after using it extensively enough to eliminate genetic risk factors, because:
1. There are always random chromosomal mutations that you might want to screen for.
2. If we get accustomed to "picking the best embryo", then we might get used to all the benefits to be had from selecting the best combination of the parents' genes, rather than just removing any actively-bad genetic risk factors. (Selecting for the highest-IQ combination being one obvious example.)
The marginal benefit is very small compared to intentionally selecting a mate for good genetics, or if you have a history of disease in your own family, not breeding at all.
Other complicating factor is the marginal benefit depends on where you sit on the margin. If you have very healthy genes already, you get a lot less out of this than someone who doesn't.
And, obviously, if you're not a woman, this isn't your choice anyway.
after some point this process will go from obscure to mainstream, and then it will be something parents desperately want to do, along the lines of getting accepted to the best pre-K daycare and enrolling kids in college-prep classes in middle school.
I would. There is an history of mental illness in my family. If IVF is the only way to seriously reduce risk, I'll go with it. The current plan is to not have any, since I don't wish the hell that can be on my children. I get why people would think that weird but I would be unable to look at my child face if there were any way to avoid a depressing disability that I didn't do.
My wife's brother is an institutionalized schizophrenic. If this had been a reasonable option when we were having our kids, I would have gone for it 10 out of 10 times. Schizophrenia is scary as hell.
How exactly is this "relative risk reduction" measured ? They say it is "extrapolated from empirical data", but I don't know what that means.
If they mean, "we implanted N embryos, half screened using our algorithm and half at random, and the screened ones had X% less prevalence of genetic diseases in real life", then I might be mildly interested. On the other hand, if they mean, "we obtained a bunch of markers from GWAS, screened all the embryos for them, and assigned a score to each embryo based on its alleles", then I'm almost completely uninterested, because human traits (outside of a few outliers) are so polygenic that you can basically calculate whatever result you want based on how you tweak your model parameters.
One way to verify it would be the following: You sequence the genome of many pairs of siblings where one got some disease and the other didn't. You use the data from only half of the pairs to construct your algorithm. Then you apply the algorithm to the other half of the pairs, and look at how often it would have picked the one who didn't get the disease.
I don't know what they actually did. It's certainly not "we implanted N embryos, half screened using our algorithm and half at random, and the screened ones had X% less prevalence of genetic diseases in real life", because that would require them to actually implant the embryos decades ago.
Yes, I understand that you're talking about training your model and then cross-validating it. The predictive value of such a process is not zero, by any means; still, based on what I've seen from GWAS before, I doubt its effectiveness in practice.
Unfortunately, we can't possibly know if this works until many more polygenically screen babies are born, then wait 70 years to see if they really live full lives with lower rates of the indicated diseases than regular babies born to the same parents. That will likely never be possible because parents won't randomly select only some babies for screening, so we'll be left with the next best comparison to babies from other parents.
As I described in the sibling comment, and Steve Hsu has also described below, we *can* possibly know if it works by sequencing the genome of pairs of siblings who are now old, and looking at whether the method would have picked the healthier one.
You can possibly know if polygenic screening in general could work. I'm saying you can only know if one specific company's product works by actually testing the results of that product.
Time to start campaigning to get lots of IVF parents to have two kids, do polygenic screening on both, but then the lab randomly selects one of the two to ignore the parents' selection and pick an embryo at random instead (this is the control group)
Think of the targeted advertising though!!! And the intricate legal arguments for like the inclusion of a kibra variant that gives you near eidetic memory but also makes it that much easier to develop PTSD
Although given how easy it is to use patented technology that's technically different but functionally/biologically not because law is so far behind on understanding biology, I'm sure the parents won't win? On the other hand Erin Brokovich sometimes happens when irl she would have had an actual scientific case doing like neuroblastoma rates in Rockford Illinois where furniture manufacturing first took off
I am thinking that now, and thanking God (or rationalist equivalent higher power) we already have 3 passably functional kids and not in the market for any more 😉
The entertainment value is diluted for me by imagining the way parents will treat children who are so disappointing. There's enough abuse already without adding genetic engineering to the mix.
Is there a genetic basis for being emotionally abusive? Would anyone sign up for it being edited out of their children?
I'm sure that the companies will be very careful not to promise too much, for exactly this reason. I doubt you can sue them if you get cancer or a rare genetic disease, for example.
B. Even a very bright designer baby, apparently capable of reading weeks after birth, won't have lived long enough to understand that those labels correspond to extremely low absolute risks.
Poor baby probably thinks all babies are killed by consumer products by the age of 2.
I saw a search result that stated that 'dawn' is _one_ of the meanings of 'aurea' and results with other unrelated meanings as well, so I'm not so sure that 'aurea' as 'dawn' is definitely wrong, especially given both the way language works in general and that there are lots of languages (and dialects) based on the supposed origin language of the term (Latin).
And 'golden dawn' is a pretty common phrase too, 'aurea' seems (based on my cursory skim of search results for "aurea") based on Aurora anyways (which was the goddess of the dawn), so ... I'll give Scott a pass on this 'mistake'!
The adjective 'aureus' means something closer to 'the light of day,' which itself is golden - alternatively, it could mean something like shining, excellent, etc. (the etymology being derived from 'aurum')
Dawn, as in the morning/daybreak, is always aurora. It gets confusing when you factor in the unrelated 'aura -ae,' since this has a Virgilian connection with the heavens and daylight, versus the darkness of the underworld.
I am concerned that we do not understand this well enough. What if, for instance, we find that genes that correlate with disease also correlate with some particular emotion? Or that (the genes we believe increase) intelligence are inversely correlated with empathy? Under a host of horrific hypotheticals, widespread adoption could be catastrophic, creating an entire generation that is missing an attribute, has far different statistical outcomes from the norm, etc. I leave it to those with more imagination than myself to generate better scenarios, and those with more knowledge to figure out if they could actually happen.
This makes Heinlein's Methuselah Project look like it belongs in a high school science fair. This is strong medicine. It will be applied exclusively by the ignorant (since that's all of us, right now) and, also, mostly by idiots who haven't thought about what they are doing.
I strongly suspect it will be done, at least in some if not many places, and I believe the impact will be large. I do not know if it will be positive or negative - or, more likely, both.
"For He knows what we are made of, and that it is dust." But perhaps not for much longer.
Chin up. Biochemistry is so unbelievably stuffed with redundancy it seems very unlikely to me to be anything more impactful than the cloning of Dolly the sheep, which was going to usher in a Brave New World of cloning in about 8 months flat, 25 years ago now. If it were *possible* to routinely pick out 1 out of 10 IVF embryos that would end up very different from its parents -- 2x smarter, wholly free of disease, able to leap buildings in a single bound -- then this kind of thing would happen out in the wild in 10% of natural births already -- and that is not what we see. Children generally end up pretty similar to their parents, and wild deviations, where a genius is born from IQ 105 parents, say, are exceedingly rare, 1 out of 10 million kind of events. So yeah if you could screen 10 million embryos, you might pick out the one rara avis and get results that would make the Howard Foundation proud.
It's literally just embryo selection. Genetic counselors have been informally doing this for a while. They're not interfering or engineering in any way- like editing blastocysts with strong promoters around magic machiavellian genes to ensure psycho hitler.
Like we are literally closer to Jurassic Park being a reality than your vision of human genetics - https://pleistocenepark.ru
- the literal Adolf Hitler was conceived naturally
- psychopaths are conceived naturally every day
- a small percentage of people are willing to defy norms and start respectability cascades
It is possible to set everything on fire with only quantitative changes.
That said, there are some obvious bounds here. If a trait has frequency X0 in the population, this form of selection is taken up at frequency Y, and the average number of embryos selected from is Z, then the frequency of the trait X1 after one generation is bounded by:
Yeah, this is right. In particular (1) I bet those graphs reflect prediction from pooled population samples, not causality estimated from sibling groups and leveraging the "lottery of meiosis". In which case, embryo selection may be less effective than predicted, maybe even have zero effect in some cases. But also (2) we have no idea what all these genes actually do, we've got no idea about the biological pathways between them and the outcome, we've got no idea of possible side effects. Basically, this might be about as irresponsible as handing out an untested drug.
O brave new world, that has such people in 't!
Hug me, 'till you drug me, honey; kiss me, 'till I'm in a coma!
https://www.nejm.org/doi/full/10.1056/NEJMsr2105065?query=recirc_inIssue_bottom_article
Selecting for intelligence feels likely to be banned (unfortunately as this could be hugely beneficial for humanity if widespread) but in theory selecting against disease should also be pro intelligence as both are correlated through mutational load, right?
Given the prices involved here and methods used, this might be...really difficult to ban? If you let parents see each embryo's data before they make a choice, and the information necessary to make that selection is on SciHub, then, uh, *wink wink*...
Would be hard to ban any use of this method considering how many diseases it can fight (and how sympathetic people suffering from these diseases are), and I think any use of this method quickly results in selection for whatever parents want (whether that's intelligence or aesthetic traits).
Yeah, I agree. In fact, I don't know if you even need the data. Here's an article I saw linked over at marginal revolution a while back:
"Children resulting from frozen embryos were more socially adept than those implanted fresh after eggs were fertilised. The children also moved better, had superior communication skills and showed more independence. Allan Pacey, a fertility expert at the University of Sheffield, suggested that this was caused in part by the rigours of the thawing process. Not all embryos survive thawing, and perhaps those that do are “stronger”, he said.[11] Perhaps freezing and thawing embryos is an inadvertent eugenics process, most successful with embryos having a low mutation load."
https://quadrant.org.au/magazine/2015/05/eugenics-ready/
There does seem to be a selection process, post implantation.
"Good obstetric outcome (GBO), defined as a singleton, term, live birth with appropriate for gestational age birth weight, was the primary outcome measure. Secondary outcomes included live birth, clinical pregnancy, spontaneous abortion, preterm birth, multiple births and gestational age-adjusted weight. Outcomes were modeled using the generalized estimating equation approach.
MAIN RESULTS AND THE ROLE OF CHANCE
"Overall, fresh transfer was more likely to result in a live birth (55.7% versus 39.5%; adjusted risk ratio 1.21; 95% CI 1.18–1.26; P < 0.001). "
https://academic.oup.com/humrep/article-abstract/35/12/2850/5982008?redirectedFrom=fulltext
Interesting; so it sounds like they found the opposite: fresh beats frozen.
Well, I'm not sure they're not entirely "opposite" results. If freezing meant embryos with problems were less likely to develop to term that would explain the improved health outcomes from frozen infants.
I'm saying that in contrast or addition to the Marginal Revolution advanced hypothesis "Not all embryos survive thawing" it could be that thawed embryos with health issues are less likely to be delivered to term.
If the observations are all accurate then I really wonder what mechanism mediates this.
That makes me wonder-- is freezing merely a mechanism to improve outcomes for IVF, or can it be used to improve outcomes even relative to non-IVF natural conception?
That is, would it make sense for couples to freeze embryos specifically to filter out worse ones (aside from the usage of PGS)?
If it's banned in the US but legal in, say, Brazil, then presumably a lot of wealthy ambitious couples will go on a long vacation (or several short ones) in Brazil and come back with the woman pregnant, and regulators in the US unable to do anything about it.
Citizen jurisdiction. They might not be able to force the woman to have an abortion ("cruel and unusual"), but a country can absolutely throw its citizens in prison for what they did abroad.
(The USA in particular also has some really dodgy long-arm statutes which give them jurisdiction if any money touched the US banking system at any point.)
If they can find it out.
I hope the Overton window shifts enough that selecting for intelligence gets accepted -- I always think of this passage by Nick Bostrom in https://www.nickbostrom.com/views/science.pdf:
"There are three ways to contribute to scientific progress. The direct way is to conduct a good scientific study and publish the results. The indirect way is to help others make a direct contribution. ... A third approach is to marry the first two and make a scientific advance that itself expedites scientific advances.
A “superficial” contribution that facilitates work across a wide range of domains can be worth much more than a relatively “profound” contribution limited to one narrow field, just as a lake can contain a lot more water than a well, even if the well is deeper.
No contribution would be more generally applicable than one that improves the performance of the human brain. Much more effort ought to be devoted to the development of techniques for cognitive enhancement, be they drugs to improve concentration, mental energy, and memory, or nutritional enrichments of infant formula to optimize brain development.
Imagine a researcher invented an inexpensive drug which was completely safe and which improved all‐round cognitive performance by just 1%. The gain would hardly be noticeable in a single individual. But if the 10 million scientists in the world all benefited from the drug the inventor would increase the rate of scientific progress by roughly the same amount as adding 100,000 new scientists. Each year the invention would amount to an indirect contribution equal to 100,000 times what the average scientist contributes. Even an Einstein or a Darwin at the peak of their powers could not make such a great impact. Meanwhile others too could benefit from being able to think better, including engineers, school children, accountants, and politicians."
"Imagine a researcher invented an inexpensive drug which was completely safe and which improved all‐round cognitive performance by just 1%"
Sounds like modafinil to me
Nah, that's temporary. A better comparison is salt iodization: https://www.givingwhatwecan.org/post/2016/01/are-we-underestimating-benefits-salt-iodization/
But interventions like salt iodization only help better realize potential, not increase it. It's the latter I'm curious about. Unfortunately this is a really really hard problem: https://www.gwern.net/Drug-heuristics
Practically, what's the difference between "realizing" potential and "increasing" it? Is it the upper bound of what's possible? I don't know if we know what that upper bound looks like for polygenic testing for either a given individual or a population.
Yes, the main difference is about the upper bound. Addressing iodine deficiencies (from the limited reading I've done) is likely to be a very cost-effective method to improve cognitive ability in the developing world. I think it's a shame we don't prioritize these types of low-cost interventions that have such huge long-term effects.
In regards to your second question, the upper bound for polygenic testing for a single generation is about 1 standard deviation for any given trait given current limitations on oocyte production and GWAS sample sizes. For most diseases it's closer to 0.3-0.5 standard deviations.
For multiple generations the limits are much much higher. For example, Steven Hsu has estimated that if you naively add up all the alleles that contribute positively to IQ, additive variance alone would predict that genome would have an IQ of about 1000. Obviously in practice that wouldn't work even if you could do it because at some point the relationship between positive alleles and phenotype would break down. But it gives you an idea of what might be possible in the future.
Further evidence can be found in animal and crop breeding programs that have been going on for thousands of years. We have increased the oil and protein content of the ancestors of modern corn by 30 standard deviations over the last 10,000 years. And while humans are not corn, many of the genetically influenced traits we care about (like disease risk) are controlled by a similarly large number of genes (meaning they are probably capable of undergoing similarly large increases with enough selection pressure).
"For example, Steven Hsu has estimated that if you naively add up all the alleles that contribute positively to IQ, additive variance alone would predict that genome would have an IQ of about 1000."
We seem to agree that that wouldn't actually work. So doesn't that indicate a problem with the model we're using? We do agree that not all improvements are additive, right? Some may just be different ways of accomplishing the same thing.
I feel like it's hard to predict whether this technique could produce people who were that much faster, better, or stronger than the best currently existing humans without some kind of tradeoff. Eventually we'd presumably start running into issues related to cranial capacity. Or trading off generalized for specialized intelligence.
The notion of "+.2 IQ points for this gene with no consequences or costs" just seems deceptively ... neat and predictable.
"Further evidence can be found in animal and crop breeding programs that have been going on for thousands of years. "
Sure. And the Green Revolution has accelerated that improvement with fertilizers, pesticides, watering, etc. It's also produced plants that don't survive well outside of a farm. It's produced plants that serve humans rather than themselves (since obviously wild plants have had millions of years to improve themselves for their own purposes.) Whats the equivalent of applying that metaphor to humans? Being able to use more fertilizer and water, as provided by farmers, is win-win. Maybe human intelligence can do something similar. More food = more cholesterol for neurogenesis? But then, we've demonized cholesterol and we give people drugs so they produce less of it so they can live longer. At some point crop improvements are going to have to hit some kind of physical limit in terms of photosynthetic efficiency and available light. Even if per hectare crop production is not yet leveling off for some crops. Monoculture offers a benefit in terms of being able to harvest ripe crops at a set time, reducing poaching by animals. But the loss of genetic diversity also increases vulnerability to pathogens. Genetic diversity exists for a reason. IQ positively correlates with alcoholism, for example.
An endurance runner has more slow twitch muscles. A sprinter has more fast twitch muscles. Once we eliminate various physiological defects, further improvements are tradeoffs. And I don't know how to be certain that this wouldn't be an issue with polygenic testing.
Polygenic testing is a very exciting development. It can help address things like increasing mutational load in an ethical fashion. But we're dealing with self modifying systems whose behavior we can't fully predict or understand. That seems like a setup for surprises.
I think talking about Overton windows is rather premature at this point, as there's no hard evidence that embryos can be effectively screened for intelligence at all. Merely assigning a numeric score based on a bunch of markers isn't very reliable for an extremely polygenic trait such as intelligence... especially when there's still active debate going on regarding what proportion of IQ is genetic; and even to what extent IQ correlates to real-world outcomes; or whether it even makes sense to use a single scalar quantity to answer such questions.
It’s 85 percent heritable (g), not iq scores. Especially at adult-age. Less suppressed variation due to being forced to go to school or parental choices. Cumulative exponential effects from learning faster and gaining skills faster. Same reason why bad diet in NKorea attenuates and reduces intragroup variation. Phenotypic variation is maximized and potential is realized when opportunities are maximized; sex-linked behaviours, career choices, intelligence, anything. Breeding dogs and animals have exact morphological similarities as hybridization of haplotypes or human breeding groups, it’s just politicized and suppressed because the repercussions are too big to be accepted (non-conscientious, psychopathic, low intelligence being a feedforward loop of more undesirable societal outcomes and offspring). And au contraries to the many individual variant effect hypothesis, it’s mainly few main progenitor genes with integrated mechanisms that are deterministic of neurological/biophysical health that are conserved with additional improvements made by a combination of rare and minute pathways that increase the efficiency/differentiation of brain tissue/change of rate of growth and thickening which increases the magnitude of intelligence gains. Not just a series of 0.00012 percentile variants that improve myelination or conductivity. dendrite dispersibility. Scalar measurements are invariant with respect to any and all cognitive tasks. More instructions, more complexity is inversely proportional to success rates; that includes taking medicine, making mistakes, accidents or anything. Statistical artefacts are also irrelevant and impertinent. China will be dominating and probably Israel too. The elite population has already been practicing eugenics discretely with IVF reiterated embryo selection and cross marriages stratified only by high conscientiousness, agreeableness, intellect and desire for dominance.
Relevantly Israel is leading the world in IVF so they will probably be the leader in embryo selection: https://www.nytimes.com/2011/07/18/world/middleeast/18israel.html
Judaism is bioethically permissive (abortion is legal in Israel, no one cares) and there isn't much wokeness, and there is a national defense reason to add +10 IQ to everyone so I expect them to be first or among the first to do embryo selection.
I did see an opinion poll of China; they're not actually for it. See https://www.pewresearch.org/science/2018/07/26/public-views-of-gene-editing-for-babies-depend-on-how-it-would-be-used/ and https://www.pewresearch.org/science/2020/12/10/biotechnology-research-viewed-with-caution-globally-but-most-support-gene-editing-for-babies-to-treat-disease/
https://www.theguardian.com/world/2019/dec/30/gene-editing-chinese-scientist-he-jiankui-jailed-three-years this also happened in China so I wouldn't count on it and most Taiwanese/Singaporeans are against embryo selection for intelligence (>60% against, ~30% for) so I don't think it's popular in China.
You don't have to understand all of the genes in order to select for higher intelligence. It may be extremely polygenetic, but we can still identify the gene variants with the largest effects and select for those. I think the Overton window is very relevant, since we're going to have the technology to attempt it long before we know whether it's actually possible.
To know if we can select for intelligence, we'd basically need to just do it on a massive scale in humans, then compare the IQs of children selected for intelligence to children not selected for intelligence (and obviously account for the many, many confounders). The question is whether we allow this experiment to take place or not.
> It may be extremely polygenetic, but we can still identify the gene variants with the largest effects and select for those.
As of today, we really can't do that, because intelligence is not just a linear combination of a bunch of genes. Like any polygenic trait, it's a complex network of interactions between genes and regulatory regions, all of which upregulate something while downregulating something at the same time. So assigning an IQ score based on a bunch of markers is not a simple matter of multiplying their association scores together. It's a search in multi-dimensional space, and the problem with brute statistical models -- yes, even ML models -- is that if you have enough dimensions, you have enough degrees of freedom to pretty much find anything you want.
As I said in my comment above, if someone had a model that provably selects embryos who then grow up into children who reliably perform better in the real world than controls, then I would be impressed. But merely throwing GWAS at the wall to see if it sticks is less interesting to me.
> Like any polygenic trait, it's a complex network of interactions between genes and regulatory regions, all of which upregulate something while downregulating something at the same time.
It's true of course that the genetics of intelligence are complicated. But you're incorrect in saying that this makes it impossible to select for. Most of the variance in highly polygenic traits can be explained with linear effect models. See this paper for a more thorough analysis: https://www.biorxiv.org/content/10.1101/2020.02.12.946608v1.full
> if you have enough dimensions, you have enough degrees of freedom to pretty much find anything you want.
Only if you use a weak statistical threshold for significance. But most modern genome-wide association studies use a very strict significance threshold of 5*10^-8, at which point you don't get many false positives.
> But merely throwing GWAS at the wall to see if it sticks is less interesting to me.
This is not an accurate description of how polygenic screening is done. All of the predictors used by companies currently doing pre-implantation screening are validated using test sets of siblings that weren't in the training data. And while the predictors used today don't capture all genetically caused variance (usually they capture around 20% of it), they do significantly better than chance at picking out the sibling who will go on to avoid developing a tested disease.
There are some limitations to these results (for example predictors of disease risk created with a training set of people with mostly European ancestry don't generalize well to non-Europeans), these limitations are essentially due to differences in the frequencies of certain alleles among populations whose common ancestors are sufficiently distant.
This in turn is caused by something called "linkage disequilibrium". When DNA from a sperm and egg crosses over during meiosis, the locations of such cross-overs are essentially random. But more often than not, base pairs on the chromosomes that are physically very close to one another on the genome will remain together after crossing over. This creates a problem for geneticists: if you're trying to figure out which letter in a protein-coding region is causing a 2% increase in risk of breast cancer, how do you know whether it's the A or the T right next to it? They're almost always inherited together due to linkage disequilibrium.
For people with recent ancestors, this doesn't have much of an effect on the prediction accuracy of these GWAS tests; the A and the T are very likely to be inherited together, so no matter which one was causing the increased risk of Breast Cancer, selecting against one will probably select against the other. But the longer ago two people last shared a common ancestor, the more likely it is that one of their ancestors had a split between those the letter that actually causes the change and the one that you've pinpointed as most likely to cause it. If that happens, your predictor becomes useless because it's incorrectly assigning causation.
Fortunately there's a solution. In the short term with the relatively small sample sizes we have today, we can restrict predictor use to populations in which they have been validated to work well. And in the longer run, GWAS need to start including genetic/phenotype data from a much larger, more genetically diverse group of people. Doing so will both improve the predictors AND make the benefits of embryo screening available to a larger group of people. And we're likely to develop many useful screening tests along the way that can be used to target health interventions at people most at-risk for certain conditions.
I don't think something has to be achievable for there to be a productive discussion about Overton windows. Five years ago self-driving cars were a novelty that still to this day don't function well enough for commercial use, but there were all sorts of ethical discussions around their decision making that helped create a societal opinion about the technology. Similar discussions take place all the time with the potential problems/benefits of AI and how we should approach the technology.
Boosting human intelligence through this type of gene selection is a bit further off than some of these things, but the potential is there for it to arrive within our lifetimes, and discussions about it will surely happen before it is fully feasible.
>Five years ago self-driving cars were a novelty that still to this day don't function well enough for commercial use, but there were all sorts of ethical discussions around their decision making that helped create a societal opinion about the technology.
I'm pretty sure that the "societal opinion about the technology" is that society expects the technology to produce cars that simply do not crash, and that nobody who matters is going to say "It's great that our cars make what we consider proper ethical decisions about who or what to crash into". The consequentialist nerds got to imagine that the Trolley Problem finally had a real-world application; everybody else said "if you're talking about what the car will decide to crash into, get back to us when you've got the car deciding not to crash".
Anything that avoids anyone mentioning the trolley car problem ever again is OK in my book
We already know polygenic scores can predict which sibling has higher intelligence and educational achievement. Predicting which embryo will is no different.
> ... A third approach is to marry the first two and make a scientific advance that itself expedites scientific advances.
I thought that was going to be "A third approach is to marry the first two so they have more scientist kids."
Strange loop is strange.
If having 100,000 more scientists made a huge difference, we would *already* have seen that effect, since there are *way* more scientists working today than at any time in human history, even on a per capita basis. If it made a giant difference, we ought to be seeing Einstein/Fermi/Newton breakthroughs on an annual basis, certainly much faster than we saw them before 1939 -- and nothing could be further from the truth.
The entire calculation is based on a dubious premise about the nature of scientific advance, id est that it proceeds like some kind of construction of an earthen dam, and if you have 200,000 peasants with buckets it gets done 2x as fast as if you had only 100,000. But so far as I can see, an examination of history suggests quite the contrary, that science advances largely by abrupt and difficult to explain (let alone predict) highly contingent paradigm shifts that are the result of some lucky coincidence of improved instrumentation, cultural readiness for a paradigm shift, and some well-prepared brilliant mind that happens to be born at the right time in the right place. You might say it happens by quantum transitions, like absorption of an visible photon to cross a band-gap, and in such a case it doesn't really matter how many IR photons with energy much less than the band gap you provide, nothing happens until the big hv comes along.
> If having 100,000 more scientists made a huge difference
I think you're thinking of "huge" in relative terms, while Bostrom (and I) think of it in absolute terms. Like how a billion dollars is a huge amount of money, but negligible vs say Apple's market cap. I agree in relative terms it's small, especially given that (to quote Scott) "constant growth rates in response to exponentially increasing inputs is the null hypothesis" (https://www.lesswrong.com/posts/v7c47vjta3mavY3QC/is-science-slowing-down). It's still a lot larger than the first 2 sorts of advances, usually.
I agree with the content of your second paragraph, but am puzzled by how it wouldn't be affected by talent pool size (whether size = number of people or "amount of talent", eliding difficulties in aggregating talent for this to be measurable).
It's affected by the talent pool size up to the limit of saturation, but not afterward. Think of enzyme kinetics: once you saturated the enzyme, you don't get any more product by boosting the substrate concentration, because the substrate concentration is no longer the limiting factor on the rate.
So the question is: have we reached saturation on the talent pool yet? That is, is everyone who *can* do scientific breakthrough thinking identified, recruited, and remunerated sufficiently that the availability of talent is no longer the limiting factor? I think a decent case can be made for "mostly yes." You don't really see much in the way of Ramanujans escaping otherwise impoverished lives by sheer luck (writing a letter to Hardy). We're pretty darn good at identifying people who have the requisite smarts and easing their way into the field if they want to be in it. (With some caveats, of course: I'm amused that the woman who was one of the pair that invented the modified-nucleoside tech used to make the mRNA COVID vaccines had a rough time at Penn and ended up leaving because people thought she was an unproductive crank, and now there are people calling for her to share the Nobel for the work.)
That begs the question of what the limiting factor might be, if it isn't the size of the talent pool. As I said below, I don't have any strong ideas on what it is, other than some vague feeling that resources are not quite distributed right, that the *most* talented of the talent pool are getting the most time/money to do their work, a the expense of the less talented. (And here I mean "talented at producing breakthrough ideas.") But I have no idea how to identify such people algorithmically, although I tend to feel I can spot them when I meet them -- it's a certain weird independence of thought, an ability to spot the weakness in the conventional wisdom. It's the person who says "this phlogiston theory is all well and good, very logical and explains a lot, but I wonder what happens if I oxidize some mercury."
Hmm, thanks for the food for thought, I think it's persuasive.
re: what the limiting factor may be, perhaps this Fast Grants postmortem is relevant -- its authors include Tyler Cowen and Patrick Collison: https://future.a16z.com/what-we-learned-doing-fast-grants/
Sample quote:
"A common theme across all of these is that fairly obvious opportunities were not pursued by incumbent institutions. We give examples of actions Fast Grants took not to indicate some kind of supposed brilliance but rather to emphasize the opposite: Fast Grants pursued low-hanging fruit and picked the most obvious bets. What was unusual about it was not any cleverness in coming up with smart things to fund, but just finding a mechanism for actually doing so. To us, this suggests that there are probably too few smart administrators in mainstream institutions trusted with flexible budgets that can be rapidly allocated without triggering significant red tape or committee-driven consensus."
For the sort of science that needs a research laboratory to be done properly, the limiting factor is probably the supply of research laboratories. Fully stocked, staffed, and maintained research laboratories are not cheap, and lots of people who aspire to be and have the education/credentials to be scientists seem to never land a position in one (except maybe as chief bottle-washer to a "real" scientist).
There's some science that can be done just by staring into a coffee cup and thinking real hard, https://www.beanthinking.org/?p=1019, or maybe poking at the keyboard of a commodity PC. But there's a great deal more science that deceptively seems like it should be this sort of science, but veers rapidly into junk if not grounded by experimental verification and model validation. For which, you need research labs, and if you exceed the research-lab limit you just wind up with lots of wannabe scientists thinking up junk science.
> I think a decent case can be made for "mostly yes."
I think it's "almost certainly not". The internal politics, incentives and funding structures around science are pretty perverse, and I think this drives a lot of talent away.
Aren't you a hard hitting outfielder for the San Diego Padres?
Shhh. I get a little bored with just hitting and throwing a ball, sometimes, but management doesn't approve of moonlighting.
I would like to challenge your central premise: that we aren't seeing more scientific development than ever before.
Medicine is the classic example. Today there are people studying and creating scientific breakthroughs on rare diseases. We lay people don't see this unless we have a relation struck by this rare condition. These are really hard, breakthrough discoveries that save lives. Are they society changing? Yes, if it's your loved one.
Without all these scientists you couldn't focus on rare diseases. To society at large, these are relatively minor and, to be honest, insignificant improvements most of the time (but can lead to breakthroughs: https://labblog.uofmhealth.org/rounds/10-studies-highlight-importance-of-rare-disease-research)
Yes we haven't learned how to increase our rate of Einsteins (not sure I agree but I'll accept it), but the breadth across disciplines and niches that we are advancing human knowledge today is unprecedented.
I'm not sure, but I don't think I agree with that. If we consider advances in the context of what already exists, and the resources available for implementation, I think a strong case can be made that there are many periods of history in which general technological advance was significantly higher -- and was reflected in a much strong rate of general social improvement. Let's say the advances in chemistry, thermodynamics, engineering, metallurgy and civil engineering in the 1600-1700 timeframe, for example. Or even the advances in physics in the 1880-1950 timeframe.
I agree biochemistry or molecular biology is having a renaissance right now, but I don't agree that's due to the number of people working in it. Those advances are largely built on factors that were invented 50 years ago -- breakthroughs in instrumentation (NMR) or technique (PCR), or brilliant insights (Watson-Crick model). The surge of people into the field in the last 20 years is I think a "Gold Rush" like *consequence* of smart people realizing the intellectual endeavors in this field might pay off better than in others, rather than the reverse.
I've been in the business (science research) a long time, and my anecdotal observation (insert the usual caveats) is that advance really is a lot like spectroscopy in quantum world: to go from Here to There, where There is some nontrivial and significant advance, it usually seems a quantum event, someone has to just take one big jump, all at once, and it simply can't be duplicated by any number of people taking any number of small jumps.
There are far more scientific papers published today than were published when I started in the business in the 1980s, but I would be hard-pressed to say that the density of genuinely transformative ideas has increased. We may have essentially hit "saturation" in the sense that throwing more (random) people and money into the process won't get us anything but more chaff, not any more wheat. It's not really the rate-limiting step any more.
What is, though? I don't know. I would vaguely say "getting money/time into the *right* hands -- of people who are capable of genuine breakthrough thinking." Problem is, they're really hard to identify. The guy who invented PCR is basically a surfer bum. There are plenty of people who had one brilliant idea and did zip before and after. It seems deeply serendipitous in many (althought not all) cases. How you improve this I don't know, although you might be able to nibble around the edges by improving the ability of people to just sit around and tinker with "dumb" (by the judgment of committees) ideas, since it's really only ideas considered at the time by the majority to be "dumb" that can turn out (in hindsight) to have been brilliant.
I can think of an avenue that isn't mentioned here but that I'd consider pretty damn important: academia is a horrid place to work at. Researchers have little choice in what they study (and I'd argue that most of the greatest advancements came from people obsessed with their field of study, whose research was guided by interest and their own ideas, not primarily where they'd find an underpaid job). Wages are laughably low. Job security is basically nonexistent unless you get tenure. If driven, intelligent, ambitious people don't get a chance to pursue their ideas, then they likely won't make the breakthrough they are capable of, and very few people really get that chance.
Anecdote: a friend of my mother's defended her thesis while undergoing chemotherapy, and lived apart from her husband and daughter for a decade because their careers were impossible to consolidate in a single city. She's a professor now, but nobody should have to go through that to have a shot at this.
How many people give up on research because of life circumstances? Not wanting to sacrifice their family, or sanity, for meager pay?
Then there's the ruthless pressure to publish, tweak data, the time spent writing grant applications etc, and the incentives that lead to large numbers of basically worthless papers, studies done on twenty college students and the like. Many of the people writing those want to do real research, but can't.
I don't have a solution here, but if we could aim the existing brainpower better (by which I mean, aim it less, because self-aimed talent is always more effective than people forced to work on stuff they're not passionate about) that might lead to more advances than increasing that brain power, or doubling the number of people possessing it.
All this (to the extent it holds) applies to fundamental science. On the margin, would additional researchers have more benefit in fundamental science (which is mostly done in academia) or at applied science (which is mostly done at private companies)?
Somebody wrote that quite some breakthroughs of past centuries were done by English clergy who had comfortable and safe living conditions, good education and enough time on their hands to pursuit whatever they liked. They had no hightech labs of course.
It's something of an argument for UBI.
Pre-singularity UBI is probably going to be too diffuse for this; you need concentrations of wealth among smart well-educated wannabe researchers. And it works better if those people can hire lab techs, for which you need people who need jobs (or at least strongly benefit from jobs, which is hard to square with a UBI generous enough to buy lab equipment).
I assume that UBI gives people slack for crowd-funding and institutions, so at least small research projects involving multiple people should happen.
Those English clergy-- did them having real but not onerous day jobs do them any good?
UBI also relates to the argument from Bret Weinstein and Heather Heyer that universities are biased towards expensive research that they can take a cut of, and there isn't enough theoretical work getting funded.
I think most of your points land but for some well-researched pushback, have a look at Alexey Guzey's piece (specific to life sciences) - https://guzey.com/how-life-sciences-actually-work/
That was a great read, thank you very much, though it doesn't come away saying "everything's fine", rather "It's shit, but it still works, somehow." Life (science), uh, finds a way.
I suspect that there's some overlap with the issue of hiring competent managers. Most bosses suck at bossing - you'd think this is something the market could figure out, but it doesn't. Maybe you could even link it to politicians. The underlying common problem is one of selection of leaders/elites and of effective allocation of limited resources, and market failures seem to be the norm, as well as failures of pretty much every other mechanism we know of. I expect smarter people than me have tried to tackle this, so I have nothing to contribute at this point.
> Selecting for intelligence feels likely to be banned
Quite possibly it will be in some jurisdictions. But if not all jurisdictions ban it, then there are going to be early adopters who want to do it, and once a significant number do, a lot of other people are going to want to, to keep up. There's also the point that if you have the raw data, they can't ban it.
Having said that, +3 IQ points is not a great deal, and for most prospective parents they could get a bigger boost than that with sperm/egg donation. Some people wouldn't want that, but others would.
You say that other people are going to want to select for intelligence to keep up with the others, but how many IVF babies are there going to be to keep up with, realistically? 1-2% of annual US births are IVF, and unless something happens to drastically increase that number, I don't see how adding 3 IQ points to 1-2% of the population (at most) is a compelling reason to unban a practice people consider unethical.
Adding 3 IQ points to the entire population may be easily worth the cost of conceiving every baby through IVF. IVF can have some complications, but they don't seem particularly serious or unsolvable. (You say "unban", but is it currently banned?)
I don't necessarily disagree that conceiving every baby through IVF would be a net positive, but how do you see that happening? I can't imagine how much it would cost to scale up IVF facilities and staff 100-fold, let alone the costs of actually operating them at that scale. Is that really worth the cost? It seems exceedingly unlikely that there aren't better ways to spend that money for the same, or more, benefit.
Things usually cost less, not more, when scaled up. (They might cost more if we are hitting resource limits, but that's unlikely to be a factor for IVF.) Assuming a generous $20,000 for the cost of IVF and genetic selection, it easily pays for itself through the lifetime of the child, even with modest assumptions about the correlation between intelligence and income. And that's not even accounting for the knock-on effects (such as smart people providing greater economic surplus to others in addition to making more money themselves, or a smarter society making better political decisions).
Who is paying thhose $20k? Many people can not afford that
In a world where you can select for intelligence, you can likely select for a number of other health/well being factors as well. Anyone with the financial means to do so would then face pressure to use IVF, and a scale up could realistically happen I think.
That world is a ways away, if it ever arrives, but I don't think its outside the realm of real possibility.
Can we acknowledge somewhere that for all this IVF you need an awful lot of women who could conceive naturally to consent to it? I wouldn’t have put myself through that to raise my kid’s IQ by three points. Maybe if I had a sufficiently horrible disease running in my family, but the assumed chance of various common diseases we do tend to get wouldn’t have prompted me to do IVF just so I could do polygenic screening.
A +3 IQ points increase across the population would be a big deal, easily a bigger deal than all issues at stake at a typical election combined.
+3 IQ points is a bigger boost in cognitive ability for someone starting from a higher base. I suspect the boost is not uniform -- there are only so many SNPs you can optimize, and smarter folks are already closer to the optimal setting.
But thinking about IQ makes this boost seem zero-sum -- after all, the number itself is relative to the population, not an absolute measure of cognitive ability, i.e., not everyone can "win" in this IQ boosting game. But suppose there was some score for the aggregate effect of those SNPs. Then we'll have an absolute measure for how much headroom we all have (how far do we have before we've optimized all the SNPs?); perhaps all of us, including the smarter ones, are laughably far away from this human ideal. But against this absolute measure (and not a relative measure like IQ is) everyone can (in principle) win, and the ultimate victory condition for the human race is that everyone has maxed out their SNP score, and everyone starts from a level playing field.
>Selecting for intelligence feels likely to be banned
Fuck, does that mean I need to go all the way over to a uncucked country?
Banning it would be an enormous tragedy for humanity.
>Selecting for intelligence feels likely to be banned
Does not feel likely to me.
The 'don't discriminate against/devalue people with low IQ' argument is something that gets heard a lot by people who fetishize IQ (us), because they represent the natural opponent 'side' to the type of person who would make the argument, and the social media ecosystem is efficient at throwing people into their natural opponents to produce content.
But I don't think any of this is something the average person thinks about, and I think the average person is pretty positive on the idea of IQ=good. I don't think it would occur to enough normal people that this might be something you would want to ban, certainly not enough to result in regulation.
Well in the US maybe. But some other country will allow it and own the world in a generation or two, and everyone in the US will [still] be angry and miserable.
"this could be hugely beneficial for humanity if widespread" why?
If they're doing the whole thing for $1400, then they're not doing full sequencing on all the embryos (unless there have been some huge breakthroughs which I missed). They're probably using rna microprobes like 23&me does. This means the ability to select for traits they didn't have in mind will be limited.
So how much would it cost to sequence a whole genome? And what's the likelihood of the price going down?
Weirdly enough, some companies are claiming prices below $1000:
https://www.cnbc.com/2019/07/01/for-600-veritas-genetics-sequences-6point4-billion-letters-of-your-dna.html
But you'd have to do it for all the embryo candidates, right? So, if we take the price of $600 from the article, that's $6,000 for 10 embryos.
Scott did say it is $1400 per embryo in the article.
He said the cost was $1400, plus some extra per embryo (so the per-embryo cost is lower than that).
No, they're certainly not sequencing the entire genome. They're just doing the same old looking for SNP fingerprints. That is presumably why the results are so poor.
Full sequencing is now much cheaper than $1400. Nebula Genomics is doing it right now for $499, Dante Labs runs yearly sales offering it for 300 euros etc., the price has really come down a lot in the last 2-3 years.
It's down to $299 for accurate full-genome sequencing at Nebula. So assuming it's $1400 + $300 per embryo, one cycle of 10 embryos will cost $12k for the basic IVF + $300*10 + $1400 = $16400
What's the value of an IQ point? In terms of income it's probably about $1500/year for adults, based on eyeballing these graphs: (https://pumpkinperson.com/2016/02/11/the-incredible-correlation-between-iq-income/)
It's not linear so it sort of depends on where you are in the distribution. I'm assuming the average IVF-user will be around 115.
Assuming that $1500/year keeps up with inflation, and our discount rate equals the 30 year treasury rate equals the long term average rate of inflation, we can just multiply $1500 by about 40 years of working to get $60,000 as the value of one IQ point.
If instead your discount rate is 4% above inflation (a reasonable forward estimate of the expected return of SPY, which is below historical averages but justified by the abnormally high PE ratio presently) and the income starts 25 years after the IVF, then the value of an IQ point comes down to about $11k, and it's still worth it to get 3 points for $15200.
But assuming the $1500/year only keeps up with inflation is too pessimistic -- it should track overall income growth, which should be 2% faster than inflation. If SPY is 4% above inflation but only 2% above income growth, we use a 2% discount rate on the $1500/year annuity from 2045-2084 to get a present value of $25k.
25k is a lower bound on the value of one IQ point because it only considers the effect on individual income and not any of the other benefits of having higher IQ, such as better health and positive externalities on the community (through better science, better businesses, better political institutions, lower crime).
Getting 3 IQ points with a present value of at least 3x25k for only $16400 is an excellent deal.
The government taxes an average of 29% of people's incomes (https://www.thebalance.com/what-the-average-american-pays-in-taxes-4768594), so from the government's point of view it should be very worth it to pay $16400 now for 29% of $75,000 worth of future earnings. And arguably the government should use a lower discount rate than individuals because its opportunity cost is paying off 30 year treasuries yielding 2% instead of buying SPY yielding 6%. 29% of 3x60k is $52k of present value of future tax revenue that the government could get from paying $16400 for the whole package of IVF and embryo selection. (Before even counting the government's reduced healthcare costs due to the disease screening part).
I think that's probably not the full cost for this purpose.
(disclaimer - I've done some genome sequencing, but only of bacteria.)
The $299 nebula deal is for 30-fold coverage - that's an average, so some regions may be missing. I can't see which sequencing technology they're using, which affects what how much coverage you'd want, and also what sort of artifacts you'd expect to see. All these human sequencing packages only work because there's a good reference, so you might end up with reference sequence replacing a difference they couldn't get a good read of.
That only matters if that region actually differs, and you'll still catch most mutations I imagine. If you care about which alleles are together on the same chromosome, you'd really want to be using a long read technology. Using different chemistries together is good to get better sequences, but I doubt they're doing that.
Don't get me wrong, it's certainly impressive. Given that it's discounted from $1000, it's possible this price is a loss-leader using investor money to build the business, or something. But for the sake of argument, let's suppose not.
However, there's also the issue that an embryo isn't very many cells. You can afford to pinch a few for testing, but for a near-complete sequence you want at least many.
I'm not clear on how much saliva Nebula needs for that service, but it might be ... quite a lot of DNA. I found these statements on the web:
* "The median yield of DNA from a 2 ml saliva sample using Oragene is 110 µg", and
* "How much DNA does a human cell contain? A human cell contains about 6 pg of DNA.", and
* "DNA Sample Submission- Typically 100 to 1000 nanograms of DNA are required for whole genome or whole exome sequencing."
So realistically you probably want to aim for at least something like 200 pg of DNA as an input, or about 33 cells-worth - this ignores all losses in purification.
(You might think that 30 cells would be the minimum /anyway/, if you're getting 30-fold coverage. That's not necessarily the case, but if not it's probably also not complete coverage; you'd be sequencing some molecules multiple times and others none.) Anyway, I imagine these companies are using significantly more starting material than that.
So either you need to grow the cells up, or clone the DNA. Starting with one or two cells, the latter approach would probably give artifacts (missing sections, over-copied sections, sequence changes). If you care very much about the differences, you're going to have to go with the first option - which is going to cost more.
Aren't you assuming the IQ points are a certainty for the cost invested? That is, the investment is risk-free? But biology doesn't work that way. You're investing $15,000 for a certain probability distribution of outcomes with a mean of +3 IQ points. But there is certainly a range of outcomes, and what is the width of that distribution? If it's wide enough it's a pretty risky investment, since the odds are not bad that you get zip for your $$, and don't you need to do some discounting because of the increased risk?
I woudn't think so unless the value of the tails of the distribution is asymmetrical. For early adopters of this technology, I would actually expect it to work in the opposite direction: since the marginal benefit of one additional IQ point is steepest at the tails, and since the average early adopter of IVF is likely to be somewhat above average, the high tail of the distribution is likely to have more positive value than the lower tail, which is closer to average.
To be clear, does a couple doing this with a donor take up additional eggs that could have gone to somebody else, or does it use as many as "traditional" ivf? Neither this post nor the first linked article specified
Generally speaking intended parents will purchase an entire cycle’s worth of eggs from a donor and get the number they get. There are companies that pre-harvest the eggs and sell them one by one but it’s less common.
>Are these tribes based on geography? Are they based on race, ethnic origin, religion, IQ, what TV channels you watched as a kid? I don’t know.
>Some of it is certainly genetic – estimates of the genetic contribution to political association range from 0.4 to 0.6. Heritability of one’s attitudes toward gay rights range from 0.3 to 0.5, which hilariously is a little more heritable than homosexuality itself.
>(for an interesting attempt to break these down into more rigorous concepts like “traditionalism”, “authoritarianism”, and “in-group favoritism” and find the genetic loading for each see here. For an attempt to trace the specific genes involved, which mostly turn out to be NMDA receptors, see here)
(https://slatestarcodex.com/2014/09/30/i-can-tolerate-anything-except-the-outgroup/)
I can think of a few governments that might want to select babies for personality, so that citizens are compliant and non-rebellious. And i can think of a few political-moral points of view that might also want to select babies on personality, so as to get future people more likely to agree with their point of view.
The big limiter here is "if you're doing IVF anyway." I doubt anyone would opt for IVF just for the benefit of genetic selection, and I don't think medical providers do the procedure unless natural conception is unsuccessful or inadvisable.
I can see this shifting more and more over time though, if IVF becomes the ‘safest’ thing to do wrt genetic deseases.
What about a negative feedback effect? The more IVF is used to eliminate genetic risk factors, the fewer future generations have need of IVF.
Not necessarily. If you implant the embryo that has 1 copy of the cyctic fibrosis allele, the baby will be fine but your grandkids are still at risk.
That could happen, but I think it would be many generations away. Unless almost the entire population was screened this way it would take an awful long time for genetic selection to weed this out, assuming people continue to intermarry between classes of non-IVF/IVF born humans.
Maybe far in the future, after all genetic risk factors are removed from the gene pool. But it seems unlikely to me that we'd abandon this practice after using it extensively enough to eliminate genetic risk factors, because:
1. There are always random chromosomal mutations that you might want to screen for.
2. If we get accustomed to "picking the best embryo", then we might get used to all the benefits to be had from selecting the best combination of the parents' genes, rather than just removing any actively-bad genetic risk factors. (Selecting for the highest-IQ combination being one obvious example.)
I totally would. And although I am not judging others I personally feel morally responsible to do so.
The marginal benefit is very small compared to intentionally selecting a mate for good genetics, or if you have a history of disease in your own family, not breeding at all.
Other complicating factor is the marginal benefit depends on where you sit on the margin. If you have very healthy genes already, you get a lot less out of this than someone who doesn't.
And, obviously, if you're not a woman, this isn't your choice anyway.
after some point this process will go from obscure to mainstream, and then it will be something parents desperately want to do, along the lines of getting accepted to the best pre-K daycare and enrolling kids in college-prep classes in middle school.
I would. There is an history of mental illness in my family. If IVF is the only way to seriously reduce risk, I'll go with it. The current plan is to not have any, since I don't wish the hell that can be on my children. I get why people would think that weird but I would be unable to look at my child face if there were any way to avoid a depressing disability that I didn't do.
IVF is usually out of pocket. I don’t see why it would be restricted.
My wife's brother is an institutionalized schizophrenic. If this had been a reasonable option when we were having our kids, I would have gone for it 10 out of 10 times. Schizophrenia is scary as hell.
How exactly is this "relative risk reduction" measured ? They say it is "extrapolated from empirical data", but I don't know what that means.
If they mean, "we implanted N embryos, half screened using our algorithm and half at random, and the screened ones had X% less prevalence of genetic diseases in real life", then I might be mildly interested. On the other hand, if they mean, "we obtained a bunch of markers from GWAS, screened all the embryos for them, and assigned a score to each embryo based on its alleles", then I'm almost completely uninterested, because human traits (outside of a few outliers) are so polygenic that you can basically calculate whatever result you want based on how you tweak your model parameters.
One way to verify it would be the following: You sequence the genome of many pairs of siblings where one got some disease and the other didn't. You use the data from only half of the pairs to construct your algorithm. Then you apply the algorithm to the other half of the pairs, and look at how often it would have picked the one who didn't get the disease.
I don't know what they actually did. It's certainly not "we implanted N embryos, half screened using our algorithm and half at random, and the screened ones had X% less prevalence of genetic diseases in real life", because that would require them to actually implant the embryos decades ago.
Yes, I understand that you're talking about training your model and then cross-validating it. The predictive value of such a process is not zero, by any means; still, based on what I've seen from GWAS before, I doubt its effectiveness in practice.
Unfortunately, we can't possibly know if this works until many more polygenically screen babies are born, then wait 70 years to see if they really live full lives with lower rates of the indicated diseases than regular babies born to the same parents. That will likely never be possible because parents won't randomly select only some babies for screening, so we'll be left with the next best comparison to babies from other parents.
As I described in the sibling comment, and Steve Hsu has also described below, we *can* possibly know if it works by sequencing the genome of pairs of siblings who are now old, and looking at whether the method would have picked the healthier one.
You can possibly know if polygenic screening in general could work. I'm saying you can only know if one specific company's product works by actually testing the results of that product.
Time to start campaigning to get lots of IVF parents to have two kids, do polygenic screening on both, but then the lab randomly selects one of the two to ignore the parents' selection and pick an embryo at random instead (this is the control group)
Aurea means "golden" in Latin. Dawn is "aurora."
Also I'm picturing every yuppie mum hiring a lawyer for when Mike isn't 6' and Suzy isn't smart in 20 years and I hope it's entertaining
The entertainment value is diluted for me by the prospect of another bonaza for lawyers - 'Is your child dumber than you bargained for?'
Think of the targeted advertising though!!! And the intricate legal arguments for like the inclusion of a kibra variant that gives you near eidetic memory but also makes it that much easier to develop PTSD
Although given how easy it is to use patented technology that's technically different but functionally/biologically not because law is so far behind on understanding biology, I'm sure the parents won't win? On the other hand Erin Brokovich sometimes happens when irl she would have had an actual scientific case doing like neuroblastoma rates in Rockford Illinois where furniture manufacturing first took off
I am thinking that now, and thanking God (or rationalist equivalent higher power) we already have 3 passably functional kids and not in the market for any more 😉
The entertainment value is diluted for me by imagining the way parents will treat children who are so disappointing. There's enough abuse already without adding genetic engineering to the mix.
Is there a genetic basis for being emotionally abusive? Would anyone sign up for it being edited out of their children?
I'm sure that the companies will be very careful not to promise too much, for exactly this reason. I doubt you can sue them if you get cancer or a rare genetic disease, for example.
In extremely related news, the lobster genome is out and it's so fascinating!!! https://advances.sciencemag.org/content/7/26/eabe8290
Any insights into lobster hierarchies and room cleaning?
Hail Lobster.
What a coincidence; recently got contacted by some lobster uploads looking to escape their Russian research facility. Small world!
Aurea looks somewhat shocked - could it be...
a) her parents have just told her she's the first polygenically screened baby
b) she just read the label on her bedding(?) which seems to be about suffocating infants, or
c) she can't believe Scott thinks Aurea means "dawn" when she thought it meant "golden" (gotta update them priors baby 😉)
Vote now!
I vote D she just pooped her pants.
I swear babies poop their pants more often than they say gaga
Scott also asked whether or not a name is coincidence, so perhaps he is just unwell.
B. Even a very bright designer baby, apparently capable of reading weeks after birth, won't have lived long enough to understand that those labels correspond to extremely low absolute risks.
Poor baby probably thinks all babies are killed by consumer products by the age of 2.
I saw a search result that stated that 'dawn' is _one_ of the meanings of 'aurea' and results with other unrelated meanings as well, so I'm not so sure that 'aurea' as 'dawn' is definitely wrong, especially given both the way language works in general and that there are lots of languages (and dialects) based on the supposed origin language of the term (Latin).
And 'golden dawn' is a pretty common phrase too, 'aurea' seems (based on my cursory skim of search results for "aurea") based on Aurora anyways (which was the goddess of the dawn), so ... I'll give Scott a pass on this 'mistake'!
It just makes me think of Staph A. Which is unfortunate.
The adjective 'aureus' means something closer to 'the light of day,' which itself is golden - alternatively, it could mean something like shining, excellent, etc. (the etymology being derived from 'aurum')
Dawn, as in the morning/daybreak, is always aurora. It gets confusing when you factor in the unrelated 'aura -ae,' since this has a Virgilian connection with the heavens and daylight, versus the darkness of the underworld.
I am concerned that we do not understand this well enough. What if, for instance, we find that genes that correlate with disease also correlate with some particular emotion? Or that (the genes we believe increase) intelligence are inversely correlated with empathy? Under a host of horrific hypotheticals, widespread adoption could be catastrophic, creating an entire generation that is missing an attribute, has far different statistical outcomes from the norm, etc. I leave it to those with more imagination than myself to generate better scenarios, and those with more knowledge to figure out if they could actually happen.
This makes Heinlein's Methuselah Project look like it belongs in a high school science fair. This is strong medicine. It will be applied exclusively by the ignorant (since that's all of us, right now) and, also, mostly by idiots who haven't thought about what they are doing.
I strongly suspect it will be done, at least in some if not many places, and I believe the impact will be large. I do not know if it will be positive or negative - or, more likely, both.
"For He knows what we are made of, and that it is dust." But perhaps not for much longer.
Chin up. Biochemistry is so unbelievably stuffed with redundancy it seems very unlikely to me to be anything more impactful than the cloning of Dolly the sheep, which was going to usher in a Brave New World of cloning in about 8 months flat, 25 years ago now. If it were *possible* to routinely pick out 1 out of 10 IVF embryos that would end up very different from its parents -- 2x smarter, wholly free of disease, able to leap buildings in a single bound -- then this kind of thing would happen out in the wild in 10% of natural births already -- and that is not what we see. Children generally end up pretty similar to their parents, and wild deviations, where a genius is born from IQ 105 parents, say, are exceedingly rare, 1 out of 10 million kind of events. So yeah if you could screen 10 million embryos, you might pick out the one rara avis and get results that would make the Howard Foundation proud.
It's literally just embryo selection. Genetic counselors have been informally doing this for a while. They're not interfering or engineering in any way- like editing blastocysts with strong promoters around magic machiavellian genes to ensure psycho hitler.
Like we are literally closer to Jurassic Park being a reality than your vision of human genetics - https://pleistocenepark.ru
In case you have forgotten:
- the literal Adolf Hitler was conceived naturally
- psychopaths are conceived naturally every day
- a small percentage of people are willing to defy norms and start respectability cascades
It is possible to set everything on fire with only quantitative changes.
That said, there are some obvious bounds here. If a trait has frequency X0 in the population, this form of selection is taken up at frequency Y, and the average number of embryos selected from is Z, then the frequency of the trait X1 after one generation is bounded by:
(1-Y)*X0 + Y*X0^Z < X1 < (1-Y)*X0 + Y*(1-(1-X0)^Z)
Yeah, this is right. In particular (1) I bet those graphs reflect prediction from pooled population samples, not causality estimated from sibling groups and leveraging the "lottery of meiosis". In which case, embryo selection may be less effective than predicted, maybe even have zero effect in some cases. But also (2) we have no idea what all these genes actually do, we've got no idea about the biological pathways between them and the outcome, we've got no idea of possible side effects. Basically, this might be about as irresponsible as handing out an untested drug.
But, but, we could eliminate obesity! And everyone would be tall!